# Extrapolating retrosynthesis models This project aims to test the extrapolation cability of template-free retrosynthesis models against template-based models. ## Developer Shuan Chen and 최종휘 ## Prepare (if you don't have NVIDIA GPU avaiable) 0. Make an account in GPU-available server (P100): * Ask 박동민 to create an account for you. * Learn how to use putty or similar software to run Linux system * Install Anaconda and python ## To do (종휘) 1. Implement the existing methods from github (train and test the model): * [Molecular Transformer](https://github.com/pschwllr/MolecularTransformer) * [Chemformer](https://github.com/MolecularAI/Chemformer) * [MEGAN](https://github.com/molecule-one/megan) * [GraphRetro](https://github.com/vsomnath/graphretro) 2. Train and test these models with new dataset: * I (Shuan) will give you the dataset once you finish step 2. 3. Compare and analyze the results (with LocalRetro) ## Note 1. There may be several software (e.g. pytorch version) and hardware (gpu cuda version) conflicts. Try to find solutions using google or ChatGPT 2. The input files and method of each different model may be different. Make sure to make the input files properly.